Contents lists available at ScienceDirect

Remote Sensing of Environment

journal homepage: www.elsevier.com/locate/rse

Landsat 8 TIRS-derived relative temperature and thermal heterogeneitypredict winter bird species richness patterns across the conterminous UnitedStates

Paul R. Elsena,b,∗, Laura S. Farwella, Anna M. Pidgeona, Volker C. Radeloffa

a SILVIS Lab, Department of Forest and Wildlife Ecology, University of Wisconsin-Madison, Madison, WI, 53706, USAbWildlife Conservation Society, Bronx, NY, 10460, USA

A R T I C L E I N F O

Keywords:BiodiversityBreeding bird survey (BBS)Thermal refugiaRemote sensingConservationClimate changeThermal infrared sensor

A B S T R A C T

The thermal environment limits species ranges through its influence on physiology and resource distributionsand thus affects species richness patterns over broad spatial scales. Understanding how temperature drivesspecies richness patterns is particularly important in the context of global change and for effective conservationplanning. Landsat 8's Thermal Infrared Sensor (TIRS) allows direct mapping of temperature at moderate spatialresolutions (100m, downscaled by the USGS to 30m), overcoming limitations inherent in coarse interpolatedweather station data that poorly capture fine-scale temperature patterns over broad areas. TIRS data thus offerthe unique opportunity to understand how the thermal environment influences species richness patterns. Ouraim was to develop and assess the ability of TIRS-based temperature metrics to predict patterns of winter birdrichness across the conterminous United States during winter, a period of marked temperature stress for birds.We used TIRS data from 2013-2018 to derive metrics of relative temperature and intra-seasonal thermal het-erogeneity. To quantify winter bird richness across the conterminous US, we tabulated the richness only forresident bird species, i.e., those species that do not move between the winter and breeding seasons, from theNorth American Breeding Bird Survey, the most extensive survey of birds in the US. We expected that relativetemperature and thermal heterogeneity would have strong positive associations with winter bird richness be-cause colder temperatures heighten temperature stress for birds, and thermal heterogeneity is a proxy forthermal niches and potential thermal refugia that can support more species. We further expected that both thestrength of the effects and the relative importance of these variables would be greater for species with greaterclimate sensitivity, such as small-bodied species and climate-threatened species (i.e., those with large dis-crepancies between their current and future distributions following projected climate change). Consistent withour predictions, relative temperature and thermal heterogeneity strongly positively influenced winter birdrichness patterns, with statistical models explaining 37.3% of the variance in resident bird richness. Thermalheterogeneity was the strongest predictor of small-bodied and climate-threatened species in our models, whereasrelative temperature was the strongest predictor of large-bodied and climate-stable species. Our results de-monstrate the important role that the thermal environment plays in governing winter bird richness patterns andhighlight the previously underappreciated role that intra-seasonal thermal heterogeneity may have in supportinghigh winter bird species richness. Our findings thus illustrate the exciting potential for TIRS data to guideconservation planning in an era of global change.

1. Introduction

Understanding the determinants of species richness is a fundamentalgoal of ecology (Hawkins et al., 2003) and is essential to definingconservation priorities (Brooks et al., 2006) and predicting biodiversityresponses to global change (Brook et al., 2008). Many factors govern

species distributions and give rise to biodiversity patterns at broadspatial scales (Brown et al., 1996). Climatic factors have particularlystrong associations with species richness (Boucher-Lalonde et al.,2014), including that of endothermic species such as birds (Howardet al., 2018). Indeed, at continental to global scales, climatic factorsoutweigh land cover in explaining bird abundance (Howard et al.,

https://doi.org/10.1016/j.rse.2019.111514Received 12 March 2019; Received in revised form 17 October 2019; Accepted 29 October 2019

∗ Corresponding author. SILVIS Lab, Department of Forest and Wildlife Ecology, University of Wisconsin-Madison, Madison, WI, 53706, USA.E-mail address: pelsen@wcs.org (P.R. Elsen).

Remote Sensing of Environment 236 (2020) 111514

0034-4257/ © 2019 Elsevier Inc. All rights reserved.

T

2015) and richness (Davies et al., 2007).Among climatic factors, temperature plays a dominant role in

driving species richness. For example, warmer temperatures influencerichness patterns by increasing the number of metabolic niches (Clarkeand Gaston, 2006) and by reducing physiological burdens (Currie,1991; Currie et al., 2004), thereby permitting more species to occur.Ample evidence showing that species' distributions can be limited bytheir physiological tolerances of temperature supports these hypotheses(Khaliq et al., 2014; Root, 1988).

The strong influence of the thermal environment has importantconsequences for endothermic species which must maintain bodytemperature well above ambient temperature for necessary physiologicfunction during winter, a period typically characterized by low tem-peratures and heightened climatic stress on physiology (Newton, 1998;Williams et al., 2014). A prominent example of this influence is thenorthward shift of wintering bird ranges in North America followingrecent warming trends (La Sorte and Thompson, 2007), suggesting astrong link between temperature and wintering bird distributions.While some species are able to cope with climatic stress by modifyingforaging or roosting behaviors (Kwit et al., 2004), other species rely onfine-scale site occupancy decisions in regions with thermal hetero-geneity to minimize exposure to extreme hot or cold temperatures(Scheffers et al., 2013). Thus, regions with high thermal heterogeneitywould be expected to have higher species richness, and thermal het-erogeneity may be a particularly important factor driving speciesrichness at higher latitudes where species experience extreme coldconditions more frequently. This notion is supported by recent me-chanistic models explaining richness patterns in hyper-diverse birdcommunities across South America (Rangel et al., 2018). However,while recent research has highlighted annual temperature seasonalityas an important driver of bird species richness at broad scales (Howardet al., 2018), there has been no assessment of the importance of within-season thermal heterogeneity in explaining continental-scale speciesrichness patterns.

Data from the Thermal Infrared Sensor (TIRS) onboard Landsat 8now enables mapping of temperature at 100m resolution (Jimenez-Munoz et al., 2014). This is exciting when assessing thermal hetero-geneity, because remotely-sensed thermal data avoid the pitfalls asso-ciated with interpolated temperature surfaces from weather stations(Behnke et al., 2016), and provide more ecologically-relevant in-formation for modeling abundance (Pianalto and Yool, 2017) andpredicting species distributions and richness patterns (Albright et al.,2011; Deblauwe et al., 2016). In particular, the spatial resolution of theTIRS data (100m downscaled to 30m by the USGS) offers substantialadvantages for capturing spatial variability of the thermal environmentcompared to coarser resolution sensors, such as MODIS (1 km), to moreaccurately predict biodiversity patterns at broad spatial scales (Tuanmuand Jetz, 2015). Furthermore, utilizing thermal data directly in in-vestigations of biodiversity patterns overcomes the limitations of often-used proxy measures for temperature, such as elevation or topographiccomplexity, that may poorly reflect true temperature gradients and thatare correlated with other abiotic factors, thereby limiting robust in-ference (Minder et al., 2010).

Our goal was to develop new metrics of relative temperature andthermal heterogeneity during winter from Landsat 8 TIRS data forbiodiversity studies, and to assess their performance in predicting pat-terns of winter bird richness across the conterminous US. We chose theconterminous US as our spatial extent because it contains (1) a largerange of both relative temperature and thermal heterogeneity, (2) en-vironmental data necessary to isolate the unique contribution ofthermal factors (i.e., control data) at consistent spatial resolutions, and(3) a rich and extensive set of biodiversity data originating from stan-dardized, decades-long monitoring programs. We chose to focus onwinter bird communities because temperature plays a particularlystrong role in regulating bird populations during this season (Newton,1998). To determine winter bird richness patterns across the

conterminous US, we analyzed only resident bird species, i.e., those thatare present at a given location throughout the year, from the NorthAmerican Breeding Bird Survey (BBS), the most extensive survey ofbirds in the US.

We predicted that winter bird species richness would be positivelyassociated with warmer relative temperatures and greater thermalheterogeneity because colder temperatures challenge birds’ ability tomaintain their body temperature in an optimal range (Root, 1988), andgreater thermal heterogeneity can provide more niches potentiallysupporting more species (Letten et al., 2013). We further predicted thatrelative temperature and thermal heterogeneity would in particularaffect richness patterns of bird species with greater sensitivities tothermal environments, i.e., small- versus large-bodied species, andspecies under greater threat from climate change. Specifically, wepredicted that small-bodied species, which generally have lower ther-moregulatory abilities than large-bodied species (Porter and Kearney,2009), would be more sensitive to colder conditions and thus that bothrelative temperature and thermal heterogeneity would more stronglypredict their species richness patterns. We also predicted that climate-threatened species—those with large discrepancies between their cur-rent and future distributions following projected climate change(Langham et al., 2015)—would be more associated with areas of highthermal heterogeneity because these species tend to occur in moun-tainous areas that exert strong temperature gradients and are char-acterized by pronounced thermal heterogeneity (La Sorte and Jetz,2010; Sekercioglu et al., 2008).

2. Methods

2.1. Remotely sensed data

2.1.1. Landsat 8 Thermal Infrared Sensor (TIRS) dataTo develop metrics of relative temperature and thermal hetero-

geneity with which to predict winter bird richness patterns across theconterminous US, we analyzed data from the Thermal Infrared Sensor(TIRS) of Landsat 8. Launched in February 2013, the thermal imagingby the TIRS consists of two bands (Bands 10 and 11) centered on 10.9and 12 μm, respectively, that collect thermal data at 100m resolution(Jimenez-Munoz et al., 2014). The data are resampled to 30m usingcubic convolution to match the native resolution of imagery providedby the Operational Land Imager (Roy et al., 2014). The resampled TIRSimagery thus represents the highest-resolution source of remotely-sensed thermal data available for the conterminous US. Out-of-fieldstray light has led to some bias in both thermal bands, and has parti-cularly affected Band 11 (Barsi et al., 2014). We therefore used dataonly from Band 10 in our analysis.

We used imagery from Landsat 8 Surface Reflectance Tier 1, whichis atmospherically corrected and contains the TIRS Band 10 data pro-cessed to orthorectified brightness temperature. As our focus was onrelative rather than absolute temperature measurements, we did notrequire any further processing. Relative temperature can adequatelycapture thermal heterogeneity and may not suffer from biases in-troduced by further processing (Young et al., 2017). We first selected allimages collected in the conterminous US between December–Februaryof each year from 2013 to 2018 to bound the winter period. We thenused the included QA flags to mask out clouds (including cirrus clouds),cloud shadows, and water based on the CFMASK algorithm, and used astatic water mask derived from Landsat imagery (Hansen et al., 2013)to limit potential artifacts from ice. All imagery was obtained andprocessed in Google Earth Engine (GEE; ImageCollection ID LANDSAT/LC08/C01/T1_SR).

We calculated two metrics related to the thermal environment thatwe hypothesized would influence winter bird richness patterns. Thefirst variable is a metric of relative temperature (Fig. 1a), which we de-rived by first calculating the mean value within a 11×11 pixel movingwindow from each image and assigning it to the central pixel of the

P.R. Elsen, et al. Remote Sensing of Environment 236 (2020) 111514

2

window, and subsequently taking the median value of those meansacross the image stack. We chose 11×11 pixel window sizes(equivalent to 108,900 m2) in our analysis to maintain a relatively highspatial resolution that could then be summarized within each BBS routebuffer (see below), noting that the predictive performance of first-ordertexture measures on bird richness patterns are relatively robust to thechoice of window size (St-Louis et al., 2006).

The second variable is a metric of thermal heterogeneity (Fig. 1b),which we derived by first calculating the standard deviation valuewithin a 11×11 pixel moving window from each image and assigningit to the central pixel of the window, and subsequently taking themedian value of those standard deviations across the image stack. Forboth metrics, we tested alternative algorithms, such as calculating themean value of the coldest winter image, but those efforts sometimesresulted in stark temperature contrasts between adjacent Landsat paths,or resulted in large data gaps due to clouds and their shadows in thecoldest scene. Our approach as outlined above minimized these issuesand produced almost seamless composites with few data gaps (Figs. 1a,

b, S1).Our analysis thus resulted in two continuous layers of thermal

metrics across the conterminous US at a spatial resolution of 30m(Fig. 1a and b). We then took the mean value of each variable within a19.7 km radius buffer of each BBS route's centroid for further analysis,following established methodologies utilizing BBS data (Pidgeon et al.,2007; Rittenhouse et al., 2012). We chose a 19.7 km radius as it isequivalent to half the distance of a BBS route and thus encompasses anentire route. We calculated each BBS route centroid from the minimumbounding rectangle encompassing the georeferenced route.

Altogether, our data processing workflow included three importantsteps to minimize potential spatial and temporal biases in calculationsof relative temperature and thermal heterogeneity arising from varia-bility in the number of TIRS observations that were unobstructed byclouds and snow. First, by using an 11× 11 pixel window in our ana-lysis to calculate mean values to assign to the central pixel, we in-corporated information from the entire 108,900 m2 window neigh-borhood, reducing the influence of no-data pixels. Second, by taking the

Fig. 1. Maps of relative temperature (a) and thermal heterogeneity (b) for the conterminous US based on thermal satellite imagery from Landsat 8 (TIRS band 10).(c–j) Detailed relative temperature and thermal heterogeneity maps for regions of the Great Plains in Iowa (c, g); the southern Rocky Mountains (d, h); Tucson,Arizona and the surrounding Santa Catalina Mountains (e, i); and fields in the Central Valley near Fresno, California (f, j). Letters and boxes in (a) and (b) denote insetlocations in (c–j).

P.R. Elsen, et al. Remote Sensing of Environment 236 (2020) 111514

3

median pixel value across the image stack of six years of winter data (atotal of 18 months) in calculations, we minimized temporal patterns ofbias by deriving representative winter temperatures. Finally, by sum-marizing each metric by taking the mean value within BBS route buffersof 19.7 km radius (a total area of 1219 km2), we significantly reducedspatial patterns of bias by averaging temperatures derived from po-tentially different numbers of thermal observations per pixel.

2.1.2. Ancillary remotely-sensed environmental variablesWe included five additional ancillary environmental variables re-

lated to elevation, topography, and land cover that are known to in-fluence bird richness patterns at regional to continental scales as controlvariables in our analysis (Davies et al., 2007; McCain, 2009; Rahbekand Graves, 2001; Tuanmu and Jetz, 2015; van Rensburg et al., 2002).We used a void-filled version of the NASA Shuttle Radar TopographyMission (SRTM) digital elevation model at 1 arc-second (30m) re-solution (GEE Image ID USGS/SRTMGL_003), which matches the re-solution of our thermal data. Using these data, we derived a metric oftopographic heterogeneity known as the terrain ruggedness index (TRI;Riley et al., 1999), which is calculated as the square root of the sum ofsquared differences between a focal pixel and its eight neighbors (e.g.,within a 3×3 pixel moving window). To summarize the data for eachBBS route, we first calculated the TRI across the conterminous US andthen took the mean TRI value within each BBS route buffer as above.We further calculated the mean elevation within each BBS route bufferusing the SRTM data. Topographic heterogeneity and elevation areoften considered as proxies for thermal heterogeneity and relativetemperature in studies predicting species richness patterns (Davieset al., 2007; Rahbek and Graves, 2001). By including these variables inour analysis, our objective was to isolate the true effects of relativetemperature and thermal heterogeneity on bird richness, independentof temperature gradients and heterogeneity arising from topographiccomplexity and temperature change over elevation.

Many species also have distinct habitat affinities, such that thespatial distribution of land cover across the US can also influence birdrichness patterns. To control for this source of environmental varia-bility, we used the USGS National Land Cover Database (NLCD), aLandsat-based land cover classification for the continental US at aspatial resolution of 30m, again matching the spatial resolution of ourother environmental variables. We used the 2011 version of the NLCD(GEE ImageCollection ID USGS/NLCD), the most recent version avail-able, and that which most closely matched the time period of the TIRSdata (Homer et al., 2015). We focused on three natural vegetation typesthat represent broad-scale habitat types for birds: forest (combiningdeciduous, evergreen, and mixed forest classes), shrubland, and grass-land. To summarize the data for each BBS route buffer, we calculatedthree variables representing the proportion of all pixels that wereclassified as forest, as shrubland, or as grassland.

2.2. Bird survey data

We used data from both the North American Breeding Bird Survey(BBS, http://www.mbr-pwrc.usgs.gov/bbs) and the Christmas BirdCount (CBC, http://www.audubon.org/conservation/science/christmas-bird-count) to calculate winter bird richness patterns acrossthe conterminous US. The BBS is the most extensive standardizedsurvey of birds in the US (Sauer et al., 2017), consisting of an annualsurvey of breeding birds by volunteer observers along 39.4 km routes(4027 total routes). The primary limitation of the BBS for the purposesof our analysis is that bird distribution data are not collected duringwinter, which is a temporal mismatch with our predictor variables.Therefore, in order to adequately quantify bird richness patterns duringwinter, we restricted our analysis to include only those species withstable year-round distributions (i.e., resident species) and excludedshort- and long-distance migratory species. We reviewed the knownmovement patterns of each resident species included in our analysis

based on data in the Birds of North America (Rodewald, 2015). Weclassified species as ‘sedentary residents’ (i.e., species reported to notmove from the breeding grounds during winter), ‘residents exhibitinglocalized movement’ (i.e., species reported to remain close to thebreeding grounds year-round, with some localized movements), and‘residents exhibiting longer-distance movement’ (i.e., species reportedto show more extensive movement during the non-breeding season;Tables S1 and S2). The categories are not mutually exclusive, and agiven species could be assigned to multiple categories if, for example,individuals or populations of a species remain sedentary while othersare known to move (i.e., the species exhibits partial migration). How-ever, 95% of resident species included in our analysis were classifiedsolely as either ‘sedentary resident’ or ‘resident with localized move-ment’, or both, which suggests that richness patterns based on datacollected during the breeding season for these species should closelyreflect richness patterns during the winter.

We also considered an alternative, US-wide survey of winter birds,the Christmas Bird Count (CBC, http://www.audubon.org/conservation/science/christmas-bird-count), because it aligns moreclosely with the seasonal timing of our predictors, and we conducted allour analyses with both datasets. However, there are a number of biasesin the CBC data, and several advantages of the BBS that make the BBSmore suitable for our research questions, and that is why we show BBSresults in the main manuscript and CBC results in the SupplementaryMaterial. These include: (1) there are more BBS survey locations thanCBC survey locations (2743 BBS routes versus 1944 CBC circles from2013-2018; Fig. S7a; Fig. S8), resulting in 40% more richness ob-servations in our BBS-based analysis; (2) the survey extent of BBS routesis 3.75 times greater than that of CBC circles (buffered BBS routes coveran area of 3.34 million km2 in our analysis versus an area of 0.89million km2 for CBC circles), which captures more variation in ourpredictors and helps improve inferences on how they may influencerichness patterns; (3) BBS survey locations are more evenly distributedthroughout the conterminous US, with much more adequate coveragebetween California and Mississippi (Butcher et al., 1990), whereas CBCcircles are biased towards the Northeast and West Coast (Fig. S7a; Fig.S8); (4) land cover near BBS survey routes is more representative of theconterminous US, whereas the CBC circles are biased towards urbanareas: on average, the proportion of developed land within CBC circlesis 2.8 times higher than the conterminous US as a whole (https://www.mrlc.gov/data/statistics/national-land-cover-database-2016-nlcd2016-statistics-2016), and 2.3 times greater than the proportion of developedland within BBS route buffers based on summarized NLCD data withinrespective survey extents (BBS routes: mean=0.063, SD=0.07; CBCcircles: mean= 0.15, SD=0.17; Fig. S7b); and (5) BBS surveys arestandardized and survey effort is equal among routes whereas CBCsurveys differ in terms of survey effort by four orders of magnitude andcan include any number of observers who are allowed to cover as muchor as little of the CBC circle and use any method to count birds, in-cluding recording observations from bird feeders (Link et al., 2008;Figs. S7c and S7d). Unfortunately, there is no way to correct for thebiases listed under 3) and 4) above, and while we corrected for surveyeffort in our CBC-based analyses (see Supplementary Material), thosecorrections are imperfect. Therefore, we deemed the BBS data for re-sident species only, i.e., those species whose breeding distributionsclosely reflect winter distributions, to be more suitable for our studyand present our BBS-based findings in the main manuscript, and thequalitatively similar CBC-based findings in the Supplementary Material(Figs. S9–10).

During BBS surveys, the occurrence and abundance of all birdsalong a route are recorded during 3-min counts at 50 stops each spaced0.8 km apart along with ancillary data describing observer, observationdate and time, and weather conditions. Each route is surveyed once peryear between May and July, when species are typically breeding anddetectability is maximized. Observations are checked for validity andpotentially erroneous or questionable entries are flagged. To ensure

P.R. Elsen, et al. Remote Sensing of Environment 236 (2020) 111514

4

high data quality, we excluded all observations made by an observer intheir first year on a given route, which can lead to biased observations(Kendall et al., 1996), or where data quality flags indicated routes thata) were not randomly located; b) were not surveyed using the officialBBS sampling protocol; or c) did not meet the BBS weather, date, time,and route completion criteria. Such records are flagged with a RunTypeCode of ‘0’ in the BBS metadata, so we selected all records with aRunType Code of ‘1’, which indicates that the observation meets officialBBS criteria.

BBS routes are distributed throughout North America, but we re-stricted observations to the 48 conterminous states of the US (7.8million km2) to match the spatial extent of our predictor variables. Inaddition to excluding short- and long-distance migratory birds, aspreviously noted, we further excluded species associated with habitatsthat are underrepresented by BBS route locations (Bart et al., 2005),including those with freshwater, coastal, or marine distributions (in-cluding waterbirds and waders), as well as those which are not ade-quately surveyed using diurnal line transect methods, such as raptors,crepuscular species, and rare species (those with< 30 total observa-tions). These species also are closely tied to particular habitats, such aswater bodies, or can disperse beyond the extent of a BBS buffer, thus wewould not expect temperature to strongly drive their distributions at thespatial scale of our analysis. Finally, we excluded all hybrid species andobservations that were not identified to species-level by observers. Weadopted the BBS classifications of migration habit and breeding habitatgroups (available at https://www.mbr-pwrc.usgs.gov/bbs/guild/guildlst.html), in some cases supplemented by information from Au-dubon's online Guide to North American Birds (available at https://www.audubon.org/bird-guide), to inform these exclusions.

BBS routes have been surveyed since 1966, but to match the tem-poral resolution of our remotely sensed data, we further restricted birdobservations to those collected between 2013-2017 (BBS data for 2018were not yet available). We used BBS routes as our unit of analysis andtabulated resident species richness at each route as the total number ofunique species observed on the route during this time period. In total,we analyzed 91,582 observations of 108 resident species along 2743BBS routes.

We assigned each of the 108 resident species to a size class, dis-tinguishing ‘small-’ and ‘large-bodied’ birds by following the approachof distinguishing ‘small-’ and ‘large-ranged’ bird species in the ecolo-gical trait database EltonTraits (Wilman et al., 2014). We first usedbody mass information from Dunning (2008) for all 332 breedingspecies in the BBS (after excluding freshwater, marine, coastal, raptor,crepuscular, and hybrid species) to calculate an overall median bodymass (23.45 g; range=2.64–5791.37 g; SD=362.98 g). All speciesbelow and above the median body mass were assigned as small-bodiedand large-bodied, respectively, resulting in 166 small-bodied speciesand 166 large-bodied species. We removed from further considerationany species that was not among our focal set of 108 resident species,resulting in 34 small-bodied resident species and 74 large-bodied re-sident species (Table S1). We then calculated small-bodied and large-bodied species richness at each BBS route by tabulating the number ofsmall-bodied and large-bodied resident species.

We also assigned each of the 108 resident species to a climatesensitivity class, distinguishing ‘climate-stable’ and ‘climate-threatened’species according to the classification provided by Langham et al.(2015). The climate sensitivity classification is based on the degree towhich a species' current geographic distribution matches its projectedgeographic distribution under climate change. Those species that arepredicted to have at least 50% or more of their current and futureranges overlapping are assigned climate-stable, while species with<50% predicted overlap are considered climate-threatened. This resultedin 65 climate-stable resident species and 43 climate-threatened residentspecies (Table S1).

2.3. Statistical analysis

To quantify the relationship between each of our environmentalpredictors and winter bird richness, we fit a series of linear modelscontaining all combinations of our environmental variables as pre-dictors and winter bird richness as the response variable. We also in-cluded squared terms for relative temperature and elevation becausebird species richness sometimes shows non-linear relationships withthese variables, with richness peaking at intermediate temperatures andat mid-elevations (McCain, 2009). We separately fit models for each ofour five groups of species: all resident species, small-bodied residentspecies, large-bodied resident species, climate-stable resident species,and climate-threatened resident species. This resulted in a total of 512models considered for each group (2560 models total). Prior to fittingmodels, we assessed collinearity of our predictors by calculatingSpearman's correlation coefficients between all pairwise predictors (Fig.S2). The highest correlation between any two predictors was 0.78(between terrain ruggedness and thermal heterogeneity). All otherpairwise correlations were generally low (mean |correlation|= 0.23),so we retained all predictors in our model set. Furthermore, as a sec-ondary measure to ensure that predictors in models were not collinear,we calculated variance inflation factors for each predictor in fittedmodels.

The full model for each species group therefore included a total ofnine predictor variables along with an intercept (Table 1). For eachspecies group, we ranked models based on the Bayesian InformationCriterion (BIC), which penalizes over-parameterized models (Table 2),and centered and standardized all predictors to enable unbiased com-parisons of coefficients. We assessed the total explanatory power of themodels by calculating their adjusted R2, and assessed the contributionof each predictor in predicting bird richness for each group by plottingtheir effect sizes with standard errors. To further assess relative variableimportance of our predictors, we used hierarchical partitioning toevaluate the independent, joint, and total contributions of each pre-dictor to overall explained deviance (Chevan and Sutherland, 1991).We assessed potential biases arising from spatial autocorrelation of theBBS routes by calculating Moran's I and analyzing model residuals incorrelograms using 500 permutations. We performed all statisticalanalyses in R version 3.5.1, using the ‘hier.part’ (Walsh and Mac Nally,2013) and ‘ncf’ (Bjornstad, 2019) packages to perform hierarchicalportioning and calculate Moran's I, respectively.

3. Results

3.1. TIRS-derived relative temperature and thermal heterogeneity

Our TIRS-derived map of relative temperature showed strong gra-dients across the conterminous US (Fig. 1a). As expected, regions athigher latitudes and elevations were generally colder, yet we were also

Table 1Variables used in linear models to predict winter bird species richness.

Environmental variable Data source Temporalresolution

Spatial resolution

Relative temperature Landsat 8 TIRSBand 10

16-day 100m, downscaledto 30m

Relative temperaure2 Landsat 8 TIRSBand 10

16-day 100m, downscaledto 30m

Thermal heterogeneity Landsat 8 TIRSBand 10

16-day 100m, downscaledto 30m

Terrain ruggedness SRTMGL1 V003 Static 30mMean elevation SRTMGL1 V003 Static 30mMean elevation2 SRTMGL1 V003 Static 30mProportion forest NLCD 2011 Static 30mProportion shrubland NLCD 2011 Static 30mProportion grassland NLCD 2011 Static 30m

P.R. Elsen, et al. Remote Sensing of Environment 236 (2020) 111514

5

able to capture fine-scale and localized thermal dynamics. For example,warmer and cooler areas of the Great Plains were apparent in Iowa(Fig. 1c), temperature gradients were well-captured along the slopes ofthe Rocky mountains (Fig. 1d) and in the Santa Catalina Mountainsnear Tucson (Fig. 1e), and individual crop fields with varying thermalenvironments were delineated in the Central Valley of California(Fig. 1f).

Thermal heterogeneity derived from TIRS also showed strong pat-terns across the conterminous US (Fig. 1b). Thermal heterogeneity wasgenerally lower in less topographically complex regions, such as theGreat Plains (Fig. 1g), and highest in mountainous regions, such as theRocky Mountains (Fig. 1h). In addition to these broad-scale patterns,our approach captured thermal heterogeneity around city blocks inurban areas (e.g., in Tucson, Arizona; Fig. 1i) and at the edges of fields(e.g., in croplands in the Central Valley outside Fresno, California;Fig. 1j).

We found a moderate positive correlation (r=0.34) between re-lative temperature and thermal heterogeneity across the study region(Fig. S2), with strongest deviations at mid-high relative temperatures.For example, both relative temperature and thermal heterogeneity werehigh throughout much of Arizona and New Mexico (Fig. 1a and b), butin the adjacent state of Texas, thermal heterogeneity was fairly lowwhile relative temperature remained high (Fig. 1b). For both tem-perature variables, the presence of clouds in winter generally affectedthe midwestern and northeastern regions most (indicated by small datagaps in black in Fig. 1a and b), as well as areas on select mountainpeaks, but utilizing six years of data enabled the robust calculation ofour metrics for the vast majority of the conterminous US (Fig. S1).

3.2. Predicting winter bird species richness

The top-ranked model predicting overall winter bird richness in-cluded five predictors (relative temperature, relative temperature2,thermal heterogeneity, mean elevation, and proportion forest cover)and explained 37.3% of the variance (Table 2). Thermal heterogeneityhad the strongest positive influence on winter bird richness accordingto its effect size, followed by relative temperature (Fig. 2). A significantpositive relationship between winter bird richness and the squared termof relative temperature indicated that richness increases exponentiallywith relative temperature. Among our control predictors, proportion offorest cover surrounding BBS routes had a weak but significant positiveinfluence on winter bird richness. Mean elevation had a strong negativeinfluence on winter bird richness. Terrain ruggedness and proportion ofgrassland were not included in the top-ranked model, but had a sig-nificant weak-positive and weak-negative influence on winter birdrichness in similarly ranked models, respectively (Fig. S3).

Hierarchical partitioning analysis, which we used to determine re-lative variable importance and the independent versus joint contribu-tions of each predictor towards the variance explained by the model,indicated that relative temperature had the greatest independent effecton predicted winter bird richness patterns, followed by mean elevationand thermal heterogeneity, though thermal heterogeneity had a largertotal contribution than mean elevation (Fig. 3). In general, most of thecontributions were independent of the other predictors considered(Fig. 3).

Relative temperature, thermal heterogeneity, and mean elevationwere also always included in the top-ranked models for small-bodied,large-bodied, climate-stable, and climate-threatened species (Fig. 2).The top-ranked models for small-bodied and large-bodied speciesrichness explained 49.2% and 27.3% of the variance in richness pat-terns, respectively, while the top-ranked models for climate-stable andclimate-threatened species explained 57.3% and 38.9%, respectively(Table 2). For small-bodied and climate-threatened species, thermalheterogeneity had a stronger positive effect on richness than relativetemperature, consistent with our predictions. The opposite was true forlarge-bodied and climate-threatened richness patterns, such thatTa

ble2

Summaryof

simila

rly-rank

edmod

els(Δ

BIC

<4)

pred

icting

winterbird

rich

ness

forfive

differen

tspeciesgrou

ps,includ

ingco

efficien

tva

lues

forallen

vironm

entalva

riab

lesan

dmod

elfitan

dweigh

tstatistics

(abb

reviations:Int

=Intercep

t;df

=de

greesof

freedo

m;log

Lik=

loglik

elihoo

d;Wt=

weigh

t).

Speciesgrou

pMod

elrank

Int

Meanelev

Meanelev

2Te

rrainrugg

ed.

Rel.tem

pRel.tem

p2Th

erm.h

eter.

Prop

forest

Prop

grass

Prop

shrub

dflogL

ikBIC

Delta

Wt

R2 ad

j

Residen

t1

11.08

−1.52

––

1.33

0.32

1.37

0.55

––

7−

7112

.814

281.0

0.00

0.47

0.38

Residen

t2

11.09

−1.44

––

1.36

0.31

1.33

0.49

−0.20

–8

−71

09.1

1428

1.4

0.46

0.37

0.38

Residen

t3

11.09

−1.49

–0.27

1.39

0.31

1.18

0.39

−0.20

–9

−71

06.6

1428

4.4

3.41

0.08

0.38

Residen

t4

11.08

−1.57

–0.25

1.36

0.32

1.24

0.46

––

8−

7110

.614

284.6

3.60

0.08

0.38

Small-b

odied

12.98

−1.12

0.17

0.23

0.32

−0.13

0.71

0.39

−0.12

–10

−46

92.5

9464

.10.00

1.00

0.50

Large-bo

died

17.94

−0.66

––

1.02

0.43

0.53

––

–6

−62

85.9

1261

9.3

0.00

0.88

0.28

Large-bo

died

28.03

−0.54

−0.09

–1.03

0.44

0.53

––

–7

−62

83.9

1262

3.2

3.93

0.12

0.28

Clim

ate-stab

le1

8.47

−2.07

––

1.63

–0.82

––

–5

−63

65.2

1276

9.9

0.00

0.64

0.58

Clim

ate-stab

le2

8.37

−2.06

––

1.64

0.10

0.83

––

–6

−63

62.2

1277

1.9

2.05

0.23

0.58

Clim

ate-stab

le3

8.47

−2.10

–0.17

1.67

–0.69

––

–6

−63

62.7

1277

3.0

3.08

0.14

0.58

Clim

ate-threaten

ed1

2.71

0.60

––

−0.28

0.22

0.50

0.54

−0.11

–8

−50

47.3

1015

7.9

0.00

0.58

0.40

Clim

ate-threaten

ed2

2.72

0.50

––

−0.32

0.21

0.50

0.61

–0.12

8−

5048

.110

159.5

1.60

0.26

0.40

Clim

ate-threaten

ed3

2.70

0.56

––

−0.29

0.22

0.52

0.58

––

7−

5052

.510

160.4

2.49

0.17

0.39

P.R. Elsen, et al. Remote Sensing of Environment 236 (2020) 111514

6

relative temperature had the strongest positive effect on richness of anypredictor considered (Fig. 2). Mean elevation was a strong negativepredictor in top-ranked models for all species groups except climate-threatened species, which showed positive relationships with meanelevation (Fig. 2). Proportion of forest and grassland cover showedpositive and weak-negative relationships, respectively, with small-bodied and climate-threatened species (Fig. 2). Proportion of shrublandcover was never included in the top-ranked model for any speciesgroup, and terrain ruggedness was only included as a significant pre-dictor of small-bodied species richness with a weak-positive influence(Fig. 2).

Hierarchical partitioning indicated that relative temperature hadthe greatest total and independent effect on bird richness patterns oflarge-bodied and climate-stable species (Fig. 3). Thermal heterogeneityhad the greatest total and independent effect on small-bodied species,which were influenced to a slightly lesser degree by relative tempera-ture, proportion of forest cover, mean elevation, and terrain ruggedness(Fig. 3). Relative temperature, thermal heterogeneity, and mean ele-vation had similar total and independent effects on richness patterns ofclimate-threatened species (Fig. 3).

Of the 512 models we tested for predicting bird richness for each ofthe five species groups, between two and four models had ΔBIC valueswithin four of the top-ranked model of each, indicating similar levels ofsupport (Table 2). In each case, similarly ranked models based on BICexplained similar amounts of variance as the top-ranked model (max-imum difference in deviance explained=0.03%; Table 2). Predictors insimilarly ranked models had effect sizes with similar magnitudes anddirections as top-ranked models (Figs. 2, S3). Furthermore, the

magnitudes of the independent and joint contributions of predictors inthese models were largely consistent with those of the top-rankedmodel for each species group (Figs. 3, S4). Variance inflation factors(VIFs) for each term in top-ranked models for all species groupswere<5 (mean VIF=1.85; Table S3), indicating minimal collinearityof factors included in top-ranked models (O'Brien, 2007). Correlogramsof model residuals of total winter bird richness analyses indicated thatrichness among BBS routes exhibited only a small degree of spatialautocorrelation, and therefore that results were not strongly de-termined by their spatial configuration (Fig. S5).

Our BBS-based results were also largely consistent with those thatwe obtained when analyzing the CBC dataset (see SupplementaryMaterial), which included a total of 213 winter species, including short-distance migrants and winter visitors not captured by the BBS. Relativetemperature and thermal heterogeneity were again among the toppredictors for all winter species, small-bodied species, and climate-stable species. Consistent with our predictions, and with the BBS-basedresults, thermal heterogeneity had a stronger influence on small-bodiedspecies than on large-bodied species (Fig. S9). Similarly, richness ofclimate-threatened species was more strongly governed by thermalheterogeneity than relative temperature, while richness of climate-stable species was more influenced by relative temperature (Fig. S9).Overall, the effects of relative temperature and thermal heterogeneityfor these species groups were consistent both in terms of direction andmagnitude when comparing CBC- and BBS-based results. Results fromthe hierarchical partitioning analysis were also qualitatively similar forall species groups (Fig. S10), with the exception of small-bodied species,where relative temperature had a greater independent contribution

Fig. 2. Standardized effects sizes (points) with 95% confidence intervals (lines) for predictor variables in the top-ranked models of winter bird richness for all residentspecies and four subset groups (small-bodied, large-bodied, climate-stable, and climate-threatened species). Note that not all predictor variables are included in thetop-ranked model for each species group.

P.R. Elsen, et al. Remote Sensing of Environment 236 (2020) 111514

7

than thermal heterogeneity in the CBC-based results, opposite from theBBS-based results.

3.3. Spatial patterns in predicted bird richness across species groups

Mapping the predicted richness values resulting from the top-rankedmodel for the five species groups showed spatial patterns that largelymirrored observed richness patterns for each group from the BBS data(Figs. 4, S6). Deviations in predicted from observed richness for each

species group were largely random with respect to geographic location.Some underpredictions occurred along the west coast and throughoutthe southwestern US, and some overpredictions occurred in isolatedregions of the western US and in Florida for all resident, large-bodied,and climate-stable species. We observed smaller deviations for small-bodied and climate-threatened species, which showed minor under-prediction in California and the southwest and minor overprediction inthe Appalachian Mountains. Predicted richness patterns of large-bodiedand climate-stable species were most reflective of total resident species

Fig. 3. Results of hierarchical partitioning analysis showing the independent, joint, and total contributions of predictor variables included in the top-ranked modelsof winter bird richness for all resident species and four subset groups (small-bodied, large-bodied, climate-stable, and climate-threatened species) towards the totalvariance explained by each model. Note that not all predictor variables are included in the top-ranked model for each species group.

Fig. 4. Maps of observed richness (first column), predicted richness (second column), and richness difference (observed minus predicted; third column) for winterresident bird species across 2743 Breeding Bird Survey (BBS) routes in the conterminous US based on the top-ranked model. The left color ramp reflects observed andpredicted richness values, while the right color ramp reflects richness differences from observed, such that bluer tones are under-predictions and redder tones areover-predictions. See Fig. S6 for analogous maps for small-bodied, large-bodied, climate-stable, and climate-threatened species. (For interpretation of the referencesto color in this figure legend, the reader is referred to the Web version of this article.)

P.R. Elsen, et al. Remote Sensing of Environment 236 (2020) 111514

8

richness patterns during winter, while predicted climate-threatenedspecies richness showed patterns that most strongly contrasted withtotal resident species richness.

4. Discussion

4.1. Predictive performance of relative temperature and thermalheterogeneity in explaining bird richness patterns

Our remotely-sensed measurements of the thermal environmentrepresenting relative temperature and thermal heterogeneity duringwinter were the two strongest predictors of total, small-bodied, large-bodied, and climate-stable winter bird richness patterns across theconterminous US. Both predictors had significant positive effects onbird richness when controlling for environmental factors related toelevation, topography, and land cover. Relative temperature explainedmore variance than any other predictor in our models for residentspecies, large-bodied species, and climate-stable species, confirming itsrole as a dominant factor predicting winter bird richness. These resultslinking fine-scale relative temperature gradients with winter birdrichness support earlier, coarser-scale assessments showing tight asso-ciations between climatic factors and bird richness (Rahbek and Graves,2001).

Our results also point to the important role of thermal heterogeneityin driving winter bird richness patterns, particularly of small-bodiedand climate-threatened species. Inter-annual temperature variabilitycan exert strong influence on species’ range sizes (Chan et al., 2016),and regions of high temperature seasonality can filter out species thatare not adapted to cope with wide temperature fluctuations (Janzen,1967; Srinivasan et al., 2018). Yet thermal heterogeneity can act tobolster species richness by increasing the number of niches availableand by supporting species with divergent adaptations to the thermalenvironment (Letten et al., 2013). Furthermore, regions of high thermalheterogeneity within a given season can buffer against exposure totemperature extremes (Scheffers et al., 2013), which is important forspecies with poorer thermoregulatory capacity or ranges more tightlygoverned by climatic factors.

Indeed, the effect of temperature variability on species richnesspatterns depends on the spatial and temporal scale considered, in somecases tipping the balance towards higher rates of stochastic extinctionsthereby reducing species richness, while in others promoting speciationrates and stabilizing competition, thereby increasing species richness(Shurin et al., 2010). We found that thermal heterogeneity acts to in-crease bird species richness at continental scales, at least during winter,potentially by creating thermal refugia that can support a greaternumber of species. We further provide evidence that spatial patterns inthermal heterogeneity more strongly predict richness patterns of small-bodied and climate-threatened species, two species groups that arethought to be more sensitive to cold temperatures or largely governedby climatic factors. Large-bodied and climate-stable species may havephysiologies or other adaptations that enable them to occupy widertemperature gradients (Langham et al., 2015; Porter and Kearney,2009). Consequently, such species may not be governed as strongly bythermal conditions (Srinivasan et al., 2018), lessening their need toexploit thermal heterogeneity during winter. Our results favor in-creased consideration of thermal heterogeneity to more accuratelypredict broad-scale species richness patterns, particularly of speciessensitive to climate change.

4.2. Application of remotely-sensed thermal data for biodiversityconservation

More broadly, our analysis illustrates the promise of thermal datacaptured by satellite imagery to generate fine-scale thermal metricswith great utility for biodiversity conservation. Moderate-resolutionthermal data are readily available across the globe. Capitalizing on the

rapid growth of species locality information from standardized surveyslike the BBS, citizen science databases, and museum specimen collec-tions (La Sorte et al., 2018), future work could evaluate whether thepatterns we observe for wintering birds in the conterminous US arerepresentative of patterns globally or in other biogeographic realms,and in other seasons when biodiversity patterns can show stark differ-ences owing to migration (Elsen et al., 2018b; La Sorte et al., 2017).Thermal data in summer, for example, may reveal thermal refugia forbirds during heat waves.

Our predictive maps of winter bird richness differ considerably frommaps of total bird richness derived from static species range maps(Jenkins et al., 2015), making our maps valuable when identifyingpriority areas for conserving winter birds. Conservation priorities basedon temperature gradients and thermal heterogeneity represent coarse-filter strategies that aim to protect features of landscapes that promotebiodiversity (Tingley et al., 2014), which have advantages in conser-ving biodiversity under climate change (Elsen et al., 2018a; Lawleret al., 2015). Based on our analysis, such priority regions occur in Ca-lifornia, coastal Florida, Texas, and throughout much of the southeastand Appalachian Mountains given their high levels of overall winterbird richness (Figs. 4, S6). For climate-threatened species, prioritiesinclude the Great Basin, Intermountain West, Pacific Northwest, andthe Great Lakes region. Many of these same geographic regions are alsopriority regions for terrestrial vertebrates, freshwater fish, and trees,and correspond with areas where protection is largely lacking (Jenkinset al., 2015). Our results add support to calls for increased conservationattention in these threatened regions.

Another important consideration is the source of thermal hetero-geneity. While terrain ruggedness can drive thermal heterogeneity,anthropogenic activities such as suburban and agricultural expansioncan also significantly alter the thermal environment. For example, inthe tropics, agricultural lands are 7.6 C warmer than primary forest onaverage (Senior et al., 2017), and temperate forests with greater treedensities have warmer and less variable microclimates than disturbedforests (Latimer and Zuckerberg, 2016). While agricultural lands areimportant for many bird species during winter (Elsen et al., 2017;Rosenblatt and Bonter, 2018), habitat conversion can negatively impactbirds during breeding when they have different habitat requirements(Elsen et al., 2018b; Holmes, 2007). Thus, the positive influence ofthermal heterogeneity on bird richness we observed during winter maybe attenuated during breeding. Future work examining the interplay ofthermal heterogeneity with land cover variables would help furtherinform the mechanism by which thermal heterogeneity influencesspecies richness patterns.

4.3. Caveats and considerations

Though many factors limit species ranges and determine biodi-versity patterns (Sexton et al., 2009), advances in remotely-sensed dataare rapidly increasing our ability to investigate their roles directly(Estes et al., 2018; Radeloff et al., 2019). We focused here on thethermal environment, complementing other studies showing that re-motely-sensed data, such as metrics of productivity (Coops et al., 2009;Hobi et al., 2017), cloud cover (Wilson and Jetz, 2016), and texturemeasures as proxies of habitat structure (Culbert et al., 2010; Tuanmuand Jetz, 2015), are useful in predicting biodiversity patterns. Whileincorporating such variables into models would likely increase pre-dictive performance, the resulting model complexity potentially limitsinterpretability and the ability to directly test ecological hypotheses(Merow et al., 2014). Consequently, we opted to limit our input vari-ables to focus on testing the roles of relative temperature and thermalheterogeneity in explaining species richness patterns. In doing so, wehave demonstrated the profound influence of both variables on winterbird richness across the conterminous US, while providing an analyticalframework that can readily be applied to other taxonomic groups andother regions.

P.R. Elsen, et al. Remote Sensing of Environment 236 (2020) 111514

9

Secondly, the TIRS data represent surface-level temperatures, whichmay not precisely reflect the conditions experienced by birds, especiallyin forests. In winter, closed canopy forests can provide thermal buf-fering of cold temperatures by filtering radiation exchange (Norriset al., 2011) and altering microclimatic conditions in the understory(Latimer and Zuckerberg, 2016). While sub-canopy temperatures arecorrelated with interpolated temperatures from weather stations, thedegree of correlation is influenced by topography and landscape fea-tures such as the amount of urban area and forest edge (Latimer andZuckerberg, 2016). Methods of combining air temperature, digitalelevation model, and light detection-and-ranging (LiDAR) data havebeen proposed to more accurately capture microclimatic variability andthermal refugia (Lenoir et al., 2016), but such data are at present un-available over broad spatial scales. Given our continental scale analysisof thermal data at the BBS route level (i.e., averaged within a 19.7 kmbuffer), we suggest that TIRS data provide the best broadly availableapproximation of relative temperature and thermal heterogeneity.

Finally, our analysis took advantage of data from the most extensivebird monitoring program in the US, the Breeding Bird Survey (Saueret al., 2017), providing us with nearly 100,000 observations with whichto model bird richness patterns. However, because these surveys areconducted during the breeding season, they do not directly capturewinter bird communities, and our analysis excludes species that a)undergo short-distance migrations while still wintering in the US, andb) breed further north, but overwinter in the US. Our approach ofconsidering only resident species that have stable annual distributionsis hence conservative given the nature of the data, but we recognizethat true winter bird richness patterns likely differ to some degree.However, we have no reason to believe that focusing on these speciesqualitatively changed our results, given the limited number of speciesthat were potentially omitted. Indeed, analyzing data from the CBC,which contains short-distance migrants and winter visitors in additionto permanent residents, yielded qualitatively similar results (Figs. S9and S10). This independent validation gives us increased confidence inour conclusions that relative temperature and thermal heterogeneityare important predictors of winter bird richness patterns.

5. Conclusion

We show that temperature gradients and fine-scale thermal het-erogeneity strongly drive winter bird richness patterns, especially forspecies with greater sensitivities to the thermal environment. Ourfindings highlight the exciting potential of TIRS data to capture bothgradients in relative temperature and intra-seasonal thermal hetero-geneity that can be used to predict species richness patterns over broadscales. Our results have important implications for conservation plan-ning under global change and underscore the importance of protectingregions with high thermal heterogeneity. The global availability of TIRSdata enables future studies to investigate the predictive performance ofthe thermal environment on other taxonomic groups and in differentspatial and temporal contexts, and to understand how the thermal en-vironment interacts with other abiotic and biotic factors to governspecies richness. Such studies are needed to bolster the generality of ourfindings and to better understand when and where the thermal en-vironment plays pivotal roles in driving species richness patterns.

Declaration of competing interest

The authors declare that they have no known competing financialinterests or personal relationships that could have appeared to influ-ence the work reported in this paper.

Acknowledgements

This study was supported by the United States Geological Survey(USGS) Landsat Science Team (Grant Number 140G0118C0009). We

thank E. Razenkova, D. J. Gudex-Cross, B. Zuckerberg, A. Ives, andthree anonymous reviewers for helpful discussions that improved thismanuscript, and all the volunteers for collecting the BBS and CBC data.We also thank the USGS for providing the BBS data and T. Meehan fromNational Audubon Society for providing the CBC data.

Appendix A. Supplementary data

Supplementary data to this article can be found online at https://doi.org/10.1016/j.rse.2019.111514.

References

Albright, T.P., Pidgeon, A.M., Rittenhouse, C.D., Clayton, M.K., Flather, C.H., Culbert,P.D., Radeloff, V.C., 2011. Heat waves measured with MODIS land surface tem-perature data predict changes in avian community structure. Remote Sens. Environ.115 (1), 245–254. https://doi.org/10.1016/j.rse.2010.08.024.

Barsi, J., Schott, J., Hook, S., Raqueno, N., Markham, B., Radocinski, R., 2014. Landsat-8thermal infrared sensor (TIRS) vicarious radiometric calibration. Remote Sens. 6(11), 11607–11626. https://doi.org/10.3390/rs61111607.

Bart, J., Buchanan, J.B., Altman, B., 2005. Improving the breeding bird survey. In: Ralph,C.J., Rich, T.D. (Eds.), Bird Conservation Implementation and Integration in theAmericas: Proceedings of the Third International Partners in Flight Conference. USDAForest Service General Technical Report PSW-GTR-191, pp. 771–776.

Behnke, R., Vavrus, S., Allstadt, A., Albright, T., Thogmartin, W.E., Radeloff, V.C., 2016.Evaluation of downscaled, gridded climate data for the - conterminous United States.Ecol. Appl. 26 (5), 1338–1351. https://doi.org/10.1002/15-1061.

Bjornstad, O.N., 2019. ncf: spatial covariance functions. R package version 1.2-8. https://CRAN.R-project.org/package=ncf.

Boucher-Lalonde, V., Kerr, J., Currie, D., 2014. Does climate limit species richness bylimiting individual species' ranges? Proc. R. Soc. Biol. Sci. 281 (1776), 20132695.

Brook, B.W., Sodhi, N.S., Bradshaw, C.J., 2008. Synergies among extinction drivers underglobal change. Trends Ecol. Evol. 23 (8), 453–460. https://doi.org/10.1016/j.tree.2008.03.011.

Brooks, T.M., Mittermeier, R., Da Fonseca, G.A.B., Gerlach, J., Hoffmann, M., Lamoreux,J., et al., 2006. Global biodiversity conservation priorities. Science 313 (5783),58–61. https://doi.org/10.1126/science.1127609.

Brown, J., Stevens, G., Kaufman, D., 1996. The geographic range: size, shape, boundaries,and internal structure. Annu. Rev. Ecol. Systemat. 27, 597–623.

Butcher, G.S., Fuller, M.R., McAllister, L.S., Geissler, P.H., 1990. An evaluation of theChristmas Bird Count for monitoring population trends for selected species. Wildl.Soc. Bull. 18 (2), 129–134.

Chan, W.-P., Chen, I.-C., Colwell, R.K., Liu, W.-C., Huang, C.-Y., Shen, S.-F., 2016.Seasonal and daily climate variation have opposite effects on species elevationalrange size. Science 351 (6280), 1437–1439. https://doi.org/10.1126/science.aab4119.

Chevan, A., Sutherland, M., 1991. Hierarchical partitioning. Am. Stat. 45 (2), 90–96.Clarke, A., Gaston, K.J., 2006. Climate, energy and diversity. Proc. R. Soc. Biol. Sci. 273

(1599), 2257–2266. https://doi.org/10.1098/rspb.2006.3545.Coops, N.C., Waring, R.H., Wulder, M.A., Pidgeon, A.M., Radeloff, V.C., 2009. Bird di-

versity: a predictable function of satellite-derived estimates of seasonal variation incanopy light absorbance across the United States. J. Biogeogr. 36 (5), 905–918.https://doi.org/10.1111/j.1365-2699.2008.02053.x.

Culbert, P.D., Radeloff, V.C., St-Louis, V., Flather, C.H., Rittenhouse, C.D., Albright, T.P.,Pidgeon, A.M., 2010. Modeling broad-scale patterns of avian species richness acrossthe Midwestern United States with measures of satellite image texture. Remote Sens.Environ. 118, 140–150. https://doi.org/10.1016/j.rse.2011.11.004.

Currie, D.J., 1991. Energy and large-scale patterns of animal-species and plant-speciesrichness. Am. Nat. 137 (1), 27–49.

Currie, D.J., Mittelbach, G.G., Cornell, H.V., Field, R., Guégan, J.-F., Hawkins, B.A., et al.,2004. Predictions and tests of climate-based hypotheses of broad-scale variation intaxonomic richness. Ecol. Lett. 7 (12), 1121–1134. https://doi.org/10.1111/j.1461-0248.2004.00671.x.

Davies, R.G., Orme, C.D.L., Storch, D., Olson, V.A., Thomas, G.H., Ross, S.G., et al., 2007.Topography, energy and the global distribution of bird species richness. Proc. R. Soc.Biol. Sci. 274, 1189–1197. https://doi.org/10.1098/rspb.2006.0061.

Deblauwe, V., Droissart, V., Bose, R., Sonké, B., Blach-Overgaard, A., Svenning, J.-C.,et al., 2016. Remotely sensed temperature and precipitation data improve speciesdistribution modelling in the tropics. Glob. Ecol. Biogeogr. 25 (4), 443–454. https://doi.org/10.1111/geb.12426.

Dunning, J., 2008. CRC Handbook of Avian Body Masses, second ed. CRC Press, BocaRaton.

Elsen, P.R., Kalyanaraman, R., Ramesh, K., Wilcove, D.S., 2017. The importance ofagricultural lands for Himalayan birds in winter. Conserv. Biol. 31 (2), 416–426.https://doi.org/10.1111/cobi.12812.

Elsen, P.R., Monahan, W.B., Merenlender, A.M., 2018a. Global patterns of protection ofelevational gradients in mountain ranges. Proc. Natl. Acad. Sci. U.S.A. 115 (23),6004–6009. https://doi.org/10.1073/pnas.1720141115.

Elsen, P.R., Ramesh, K., Wilcove, D.S., 2018b. Conserving Himalayan birds in highlyseasonal forested and agricultural landscapes. Conserv. Biol. 32 (6), 1313–1324.https://doi.org/10.1111/cobi.13145.

Estes, L., Elsen, P.R., Treuer, T., Ahmed, L., Caylor, K., Chang, J., et al., 2018. The spatial

P.R. Elsen, et al. Remote Sensing of Environment 236 (2020) 111514

10

and temporal domains of modern ecology. Nat. Ecol. Evol. 2 (5), 819–826. https://doi.org/10.1038/s41559-018-0524-4.

Hansen, M.C., Potapov, P.V., Moore, R., Hancher, M., Turubanova, S.A., Tyukavina, A.,et al., 2013. High-resolution global maps of 21st-century forest cover change. Science342 (6160), 850–853. https://doi.org/10.1126/science.1244693.

Hawkins, B.A., Field, R., Cornell, H.V., Currie, D.J., Guégan, J.-F., Kaufman, D.M., et al.,2003. Energy, water, and broad-scale geographic patterns of species richness.Ecology 84 (12), 3105–3117. https://doi.org/10.1890/03-8006.

Hobi, M.L., Dubinin, M., Graham, C.H., Coops, N.C., Clayton, M.K., Pidgeon, A.M.,Radeloff, V.C., 2017. A comparison of Dynamic Habitat Indices derived from differentMODIS products as predictors of avian species richness. Remote Sens. Environ. 195(C), 142–152. https://doi.org/10.1016/j.rse.2017.04.018.

Holmes, R.T., 2007. Understanding population change in migratory songbirds: long-termand experimental studies of Neotropical migrants in breeding and wintering areas.Ibis 149 (Suppl. 1), 2–13. https://doi.org/10.1111/j.1474-919X.2007.00685.x.

Homer, C., Dewitz, J., Yang, L., Jin, S., Danielson, P., Xian, G., et al., 2015. Completion ofthe 2011 national land cover database for the conterminous United States - re-presenting a decade of land cover change information. Photogramm. Eng. RemoteSens. 81 (5), 345–354. https://doi.org/10.14358/PERS.81.5.345.

Howard, C., Flather, C.H., Stephens, P.A., 2018. What drives at‐risk species richness?Environmental factors are more influential than anthropogenic factors or biologicaltraits. Conserv. Lett. 20 (25), e12624–e12629. https://doi.org/10.1111/conl.12624.

Howard, C., Stephens, P.A., Pearce-Higgins, J.W., Gregory, R.D., Willis, S.G., 2015. Thedrivers of avian abundance: patterns in the relative importance of climate and landuse. Glob. Ecol. Biogeogr. 24 (11), 1249–1260. https://doi.org/10.1111/geb.12377.

Janzen, D., 1967. Why mountain passes are higher in the tropics. Am. Nat. 101 (919),233–249.

Jenkins, C.N., Van Houtan, K.S., Pimm, S.L., Sexton, J.O., 2015. US protected landsmismatch biodiversity priorities. Proc. Natl. Acad. Sci. U.S.A. 112 (16), 5081–5086.

Jimenez-Munoz, J.C., Sobrino, J.A., Skokovic, D., Mattar, C., Cristobal, J., 2014. Landsurface temperature retrieval methods from landsat-8 thermal infrared sensor data.IEEE Geosci. Remote Sens. Lett. 11 (10), 1840–1843. https://doi.org/10.1109/LGRS.2014.2312032.

Kendall, W.L., Peterjohn, B.G., Sauer, J.R., 1996. First-time observer effects in the northAmerican breeding bird survey. Auk 113 (4), 823–829. https://doi.org/10.2307/4088860.

Khaliq, I., Hof, C., Prinzinger, R., Bohning-Gaese, K., Pfenninger, M., 2014. Global var-iation in thermal tolerances and vulnerability of endotherms to climate change. Proc.R. Soc. Biol. Sci. 281 (1789). https://doi.org/10.1016/j.tree.2008.03.011.20141097–20141097.

Kwit, C., Levey, D.J., Greenberg, C.H., Pearson, S.F., McCarty, J.P., Sargent, S., 2004. Coldtemperature increases winter fruit removal rate of a bird-dispersed shrub. Oecologia139 (1), 30–34. https://doi.org/10.1007/s00442-003-1470-6.

La Sorte, F.A., Jetz, W., 2010. Projected range contractions of montane biodiversity underglobal warming. Proc. R. Soc. Biol. Sci. 277 (1699), 3401–3410. https://doi.org/10.1073/pnas.0606292104.

La Sorte, F.A., Thompson, F.R.I., 2007. Poleward shifts in winter ranges of NorthAmerican birds. Ecology 88 (7), 1803–1812. https://doi.org/10.1111/j.2005.0906-7590.04253.x.

La Sorte, F.A., Fink, D., Blancher, P.J., Rodewald, A.D., Ruiz-Gutiérrez, V., Rosenberg,K.V., et al., 2017. Global change and the distributional dynamics of migratory birdpopulations wintering in Central America. Glob. Chang. Biol. 23 (12), 5284–5296.https://doi.org/10.1111/gcb.13794.

La Sorte, F.A., Lepczyk, C.A., Burnett, J.L., Hurlbert, A.H., Tingley, M.W., Zuckerberg, B.,2018. Opportunities and challenges for big data ornithology. Condor 120 (2),414–426. https://doi.org/10.1650/CONDOR-17-206.1.

Langham, G.M., Schuetz, J.G., Distler, T., Soykan, C.U., Wilsey, C., 2015. Conservationstatus of North American birds in the face of future climate change. PLoS One 10 (9),e0135350. https://doi.org/10.1371/journal.pone.0135350.s011.

Latimer, C.E., Zuckerberg, B., 2016. Forest fragmentation alters winter microclimates andmicrorefugia in human-modified landscapes. Ecography 40 (1), 158–170. https://doi.org/10.1111/ecog.02551.

Lawler, J.J., Ackerly, D.D., Albano, C.M., Anderson, M.G., Dobrowski, S.Z., Gill, J.L.,et al., 2015. The theory behind, and the challenges of, conserving nature's stage in atime of rapid change. Conserv. Biol. 29 (3), 618–629. https://doi.org/10.1111/cobi.12505.

Lenoir, J., Hattab, T., Pierre, G., 2016. Climatic microrefugia under anthropogenic cli-mate change: implications for species redistribution. Ecography 40 (2), 253–266.https://doi.org/10.1111/ecog.02788.

Letten, A.D., Ashcroft, M.B., Keith, D.A., Gollan, J.R., Ramp, D., 2013. The importance oftemporal climate variability for spatial patterns in plant diversity. Ecography 36 (12),1341–1349. https://doi.org/10.1111/j.1600-0587.2013.00346.x.

Link, W.A., Sauer, J.R., Niven, D.K., 2008. Combining breeding bird survey and Christmascount bird data to evaluate seasonal components of population change in northernbobwhite. J. Wildl. Manag. 72 (1), 44–51.

McCain, C.M., 2009. Global analysis of bird elevational diversity. Glob. Ecol. Biogeogr. 18(3), 346–360. https://doi.org/10.1111/j.1466-8238.2008.00443.x.

Merow, C., Smith, M.J., Edwards Jr., T.C., Guisan, A., McMahon, S.M., Normand, S., et al.,2014. What do we gain from simplicity versus complexity in species distributionmodels? Ecography 37 (12), 1267–1281. https://doi.org/10.1111/ecog.00845.

Minder, J.R., Mote, P.W., Lundquist, J.D., 2010. Surface temperature lapse rates overcomplex terrain: lessons from the Cascade Mountains. J. Geophys. Res. 115 (D14),D14122. https://doi.org/10.1029/2009JD013493.

Newton, I., 1998. Population Limitation in Birds. Academic Press, San Diego, California.Norris, C., Hobson, P., Ibisch, P.L., 2011. Microclimate and vegetation function as in-

dicators of forest thermodynamic efficiency. J. Appl. Ecol. 102 (3), 562–570. https://

doi.org/10.1111/j.1365-2664.2011.02084.x.O'Brien, R.M., 2007. A caution regarding rules of thumb for variance inflation factors.

Qual. Quantity 41 (5), 673–690. https://doi.org/10.1007/s11135-006-9018-6.Pianalto, F.S., Yool, S.R., 2017. Sonoran Desert rodent abundance response to surface

temperature derived from remote sensing. J. Arid Environ. 141, 76–85. https://doi.org/10.1016/j.jaridenv.2017.02.006.

Pidgeon, A.M., Radeloff, V.C., Flather, C.H., Lepczyk, C.A., Clayton, M.K., Hawbaker, T.J.,Hammer, R.B., 2007. Associations of forest bird species richness with housing andlandscape patterns across the USA. Ecol. Appl. 17 (7), 1989–2010. https://doi.org/10.1890/06-1489.1.

Porter, W., Kearney, M., 2009. Size, shape, and the thermal niche of endotherms. Proc.Natl. Acad. Sci. U.S.A. 106, 19666–19672.

Radeloff, V.C., Dubinin, M., Coops, N.C., Allen, A.M., Brooks, T.M., Clayton, M.K., et al.,2019. The dynamic habitat indices (DHIs) from MODIS and global biodiversity.Remote Sens. Environ. 222, 204–214. https://doi.org/10.1016/j.rse.2018.12.009.

Rahbek, C., Graves, G.R., 2001. Multiscale assessment of patterns of avian species rich-ness. Proc. Natl. Acad. Sci. 98 (8), 4534–4539. https://doi.org/10.1073/pnas.071034898.

Rangel, T.F., Edwards, N.R., Holden, P.B., Diniz-Filho, J.A.F., Gosling, W.D., Coelho,M.T.P., et al., 2018. Modeling the ecology and evolution of biodiversity: biogeo-graphical cradles, museums, and graves. Science 361 (6399). https://doi.org/10.1126/science.aar5452. eaar5452–15.

Riley, S.J., DeGloria, S.D., Elliot, R., 1999. A terrain ruggedness index that quantifiestopographic heterogeneity. Intermt. J. Sci. 5 (1–4), 23–27.

Rittenhouse, C.D., Pidgeon, A.M., Albright, T.P., Culbert, P.D., Clayton, M.K., Flather,C.H., et al., 2012. Land-cover change and avian diversity in the conterminous UnitedStates. Conserv. Biol. 26 (5), 821–829. https://doi.org/10.1111/j.1523-1739.2012.01867.x.

Rodewald, P., 2015. The Birds of North America. Cornell Laboratory of Ornithology,Ithaca, NY. https://birdsna.org, Accessed date: 30 August 2019.

Root, T., 1988. Energy constraints on avian distributions and abundances. Ecology 69 (2),330–339.

Rosenblatt, C.J., Bonter, D.N., 2018. Characteristics of fields used by birds in winter inNew York. Wilson J. Ornithol. 130 (4), 924–929. https://doi.org/10.1676/1559-4491.130.4.924.

Roy, D.P., Wulder, M.A., Loveland, T.R., Woodcock, C.E., Allen, R.G., Anderson, M.C.,et al., 2014. Landsat-8: science and product vision for terrestrial global change re-search. Remote Sens. Environ. 145, 154–172. https://doi.org/10.1016/j.rse.2014.02.001.

Sauer, J.R., Pardieck, K.L., Ziolkowski Jr., D.J., Smith, A.C., Hudson, M.-A.R., Rodriguez,V., et al., 2017. The first 50 years of the north American breeding bird survey. Condor119 (3), 576–593. https://doi.org/10.1650/CONDOR-17-83.1.

Scheffers, B.R., Edwards, D.P., Diesmos, A., Williams, S.E., Evans, T.A., 2013.Microhabitats reduce animal's exposure to climate extremes. Glob. Chang. Biol. 20(2), 495–503. https://doi.org/10.1111/gcb.12439.

Sekercioglu, Ç., Schneider, S., Fay, J.P., Loarie, S.R., 2008. Climate change, elevationalrange shifts, and bird extinctions. Conserv. Biol. 22 (1), 140–150.

Senior, R.A., Hill, J.K., González del Pliego, P., Goode, L.K., Edwards, D.P., 2017. Apantropical analysis of the impacts of forest degradation and conversion on localtemperature. Ecol. Evol. 7 (19), 7897–7908. https://doi.org/10.1002/ece3.3262.

Sexton, J.P., Mcintyre, P.J., Angert, A.L., Rice, K.J., 2009. Evolution and ecology ofspecies range limits. Annu. Rev. Ecol. Evol. Systemat. 40 (1), 415–436. https://doi.org/10.1146/annurev.ecolsys.110308.120317.

Shurin, J.B., Winder, M., Adrian, R., Keller, W.B., Matthews, B., Paterson, A.M., et al.,2010. Environmental stability and lake zooplankton diversity - contrasting effects ofchemical and thermal variability. Ecol. Lett. 13 (4), 453–463. https://doi.org/10.1111/j.1461-0248.2009.01438.x.

Srinivasan, U., Elsen, P.R., Tingley, M.W., Wilcove, D.S., 2018. Temperature and com-petition interact to structure Himalayan bird communities. Proc. R. Soc. B 285(1874). https://doi.org/10.1098/rspb.2017.2593. 20172593.

St-Louis, V., Pidgeon, A.M., Radeloff, V.C., Hawbaker, T.J., Clayton, M.K., 2006. High-resolution image texture as a predictor of bird species richness. Remote Sens.Environ. 105 (4), 299–312. https://doi.org/10.1016/j.rse.2006.07.003.

Tingley, M.W., Darling, E.S., Wilcove, D.S., 2014. Fine- and coarse-filter conservationstrategies in a time of climate change. Ann. N. Y. Acad. Sci. 1322 (1), 92–109.https://doi.org/10.1111/nyas.12484.

Tuanmu, M.-N., Jetz, W., 2015. A global, remote sensing-based characterization of ter-restrial habitat heterogeneity for biodiversity and ecosystem modelling. Glob. Ecol.Biogeogr. 24 (11), 1329–1339. https://doi.org/10.1111/geb.12365.

van Rensburg, B.J., Chown, S.L., Gaston, K.J., 2002. Species richness, environmentalcorrelates, and spatial scale: a test using South African birds. Am. Nat. 159 (5),566–577. https://doi.org/10.1086/339464.

Walsh, C., Mac Nally, R., 2013. hier.part: hierarchical partitioning. R package version 1.0-4. https://CRAN.R-project.org/package=hier.part.

Williams, C.M., Henry, H.A.L., Sinclair, B.J., 2014. Cold truths: how winter drives re-sponses of terrestrial organisms to climate change. Biol. Rev. 90 (1), 214–235.https://doi.org/10.1111/brv.12105.

Wilman, H., Belmaker, J., Simpson, J., la Rosa de, C., Rivadeneira, M.M., Jetz, W., 2014.EltonTraits 1.0: species‐level foraging attributes of the world's birds and mammals.Ecology 95 (7). https://doi.org/10.1890/13-1917.1. 2027–2027.

Wilson, A.M., Jetz, W., 2016. Remotely sensed high-resolution global cloud dynamics forpredicting ecosystem and biodiversity distributions. PLoS Biol. 14 (3),e1002415–e1002420. https://doi.org/10.1371/journal.pbio.1002415.

Young, N.E., Anderson, R.S., Chignell, S.M., Vorster, A.G., Lawrence, R., Evangelista,P.H., 2017. A survival guide to Landsat preprocessing. Ecology 98 (4), 920–932.https://doi.org/10.1002/ecy.1730.

P.R. Elsen, et al. Remote Sensing of Environment 236 (2020) 111514

11